I'm currently working on LLM post-training and several open-source projects.
Website • Google Scholar • Hugging Face
-
Debias-SparseGPT: Bias-Aware Pruning for Large Language Models - EMNLP 2026 Main Conference Paper
-
The Fairness Collapse Phenomenon: Bias Amplification in Language Models Trained on Synthetic Data Paper
-
Fair-GPTQ: Bias-Aware Quantization for Large Language Models Paper · Code
-
HatePrototypes: Interpretable and Transferable Representations for Implicit and Explicit Hate Speech Detection - LREC 2026 Paper · Code
-
Histoires Morales: A French Dataset for Assessing Moral Alignment - NAACL 2025 Paper · Website
- DICE Leaderboard - open leaderboard evaluating fairness, safety, and trustworthiness of compressed language models.
- Debias-SparseGPT - bias-aware pruning for large language models.
- Awesome LLM Compression Safety - curated papers and resources on fairness, robustness, calibration, toxicity, and other safety effects of LLM compression.
- Fair-GPTQ - bias-aware quantization for large language models.
- HatePrototypes - transferable prototype representations for implicit and explicit hate-speech detection.